Global optimization of grey-box computational systems using surrogate functions and application to highly constrained oil-field operations. (9th June 2018)

• Record Type: Journal Article
• Authors: Beykal, Burcu; Boukouvala, Fani; Floudas, Christodoulos A.; Sorek, Nadav; Zalavadia, Hardikkumar; Gildin, Eduardo
• Is Part Of: Computers & chemical engineering. Volume 114(2018)
• Page Start: 99
• Page End: 110
• Publication Date: 2018-06-09
• Subjects: Derivative-free optimization -- Grey/black-box optimization -- Oilfield operations -- Oil-well control -- Waterflooding; Chemical engineering -- Data processing -- Periodicals; 660.0285
• DOI: 10.1016/j.compchemeng.2018.01.005
• Languages: English
• ISSNs: 0098-1354

Abstract

Highlights: High-performance computing is used to develop a highly parallelized algorithm for constrained grey-box optimization. The optimization of a 5-year operation of a benchmark oilfiled using water flooding is performed, using data from a rigorous simulation. Surrogate functions are used to reduce the dimensionality of the optimization space, via the parametrization of the pressure control profiles over time. Hybrid combinations of optimally selected and trained surrogate functions are used to represent the objective and constraints of the grey-box problem, which are iteratively optimized and updated using global optimization. Abstract: This work presents recent advances within the AlgoRithms for Global Optimization of coNstrAined grey-box compUTational problems (ARGONAUT) framework, developed for optimization of systems which lack analytical forms and derivatives. A new parallel version of ARGONAUT (p-ARGONAUT) is introduced to solve high dimensional problems with a large number of constraints. This development is motivated by a challenging case study, namely the operation of an oilfield using water-flooding. The objective of this case study is the maximization of the Net Present Value over a five-year time horizon by manipulating the well pressures, while satisfying a set of complicating constraints related to water-cut limitations and water handling and storage. Dimensionality reduction is performed via the parametrization of the pressure control domain, which is then followed by global optimization of the constrained grey-box system. Results are presented for multiple case studies and the performance of p-ARGONAUT is compared to existing derivative-free optimization methods.